vanyaland

6 ms·

1.
by vanyaland 2mo ago
ran 0.16 on a file with no config, it flags unsorted imports and `except Exception` by default now
2.
by vanyaland 2mo ago
llms write bad core data sync code by default, so that icloud bug tracks. merge policies and conflict resolution i still end up doing by hand
3.
by vanyaland 2mo ago
makes sense
4.
by vanyaland 2mo ago
worth noting the kotlin-lsp here is jetbrains own, basically intellij analysis engine running headless. you're just dropping the intellij gui
5.
by vanyaland 2mo ago
you lose the per-call approvals though. one script hitting ten tools is harder to gate than ten separate calls
6.
by vanyaland 2mo ago
How well does the fuzzy matcher resolve provenance text to origin sites? Curious what share of the 200k in the README needed hand correction.
7.
by vanyaland 2mo ago
Safari has used the same WebExtensions API as Chrome and Firefox since Safari 14. So is the hard part API coverage rather than three engines?
8.
by vanyaland 2mo ago
On iOS most apps link Apple's prebuilt libsqlite3, so compile-time defaults are out of reach.
9.
by vanyaland 2mo ago
thx
10.
by vanyaland 2mo ago
Your .ai/why/main.md is at 97KB and 148 entries. Does anything prune that, or does a long-lived branch keep growing?
11.
by vanyaland 2mo ago
Does the git lookup run before the trust check, or ignore it?
12.
by vanyaland 2mo ago
Worth clicking into .claude/agents/ in this repo. There are 25 agent definitions, 10 of them code-review dimensions.
13.
by vanyaland 3mo ago
thx
14.
by vanyaland 3mo ago
Does 2.0 expose explicit capability negotiation, or does it infer capabilities from the selected model?
15.
by vanyaland 4mo ago
Hey, short story behind this. At previous jobs I was building iOS apps with an embedded code editor. The only solution at the time was Highlightr, a Swift wrapper around highlight.js. It worked, but you paid for a JS bridge. Over winter bre
16.
by vanyaland 4mo ago
Hey HN - Ivan here, founding engineer at Diffmode. We built Diffmode for bootstrapped founders who are tired of hearing the same growth advice: write content, run paid ads, build a community, post on LinkedIn. That advice is not wrong. It i
17.
Show HN: Diffmode – Tool that builds custom growth tactics for bootstrapped SaaS(diffmode.app)
2 points by vanyaland 4mo ago|2 comments
18.
by vanyaland 5mo ago
I've been building a coding agent from scratch in Swift and using Claude Code, Cursor, and similar tools daily. This post distills five scaffolding lessons: - Instruction files: Anthropic recommends <200 lines for CLAUDE.md. The &qu
19.
Five Things I Learned About Making AI Coding Agents Work(ivanmagda.dev)
3 points by vanyaland 5mo ago|1 comments
20.
by vanyaland 6mo ago
This leak is actually a massive win. Now the whole community can study Claude Code’s architecture and build even better coding agents and open-source solutions.
21.
by vanyaland 6mo ago
Interesting observation on package definitions. Languages borrow good ideas from each other all the time, and the ecosystem is better for it. On the naming fair point, already renamed the CLI binary after an earlier comment here. The repo n
22.
by vanyaland 6mo ago
It's the harness/orchestration layer — the part that runs the agent loop, dispatches tool calls, and manages context.
23.
by vanyaland 6mo ago
Agreed. That's the core hypothesis behind this learning project — model is the magic, and the agent loop is just a thin, transparent wrapper around it. The goal of building it stage-by-stage was to prove you don't need a massive,
24.
by vanyaland 6mo ago
Good point, I'll rename the binary. Thanks for actually going through the repo.
25.
by vanyaland 6mo ago
Yeah, this is basically what I ran into too. I actually wrote about this in Stage 6 ( https://ivanmagda.dev/posts/s06-context-compaction/ ) I went with your option (1): once history crosses a token threshold, the ag
26.
Building a coding agent in Swift from scratch(github.com)
98 points by vanyaland 6mo ago|21 comments
27.
by vanyaland 6mo ago
For a lot of local workloads, sub-1 tok/s is useless in foreground and perfectly acceptable in background. If the choice is “this crashes” vs “this finishes overnight,” that’s still a meaningful capability jump.
28.
by vanyaland 6mo ago
I think part of the mismatch is that people are still looking for “more apps” as the output metric. A lot of the real value shows up as workflow compression instead. Internal tools, one-off automations, bespoke research flows, coding helper
29.
by vanyaland 6mo ago
The only thing that’s helped me is adding hard constraints: spec, architecture, small verifiable steps, and explicit decision logs.
30.
by vanyaland 6mo ago
agree, I also used Ralph loop /umputun/ralphex on GitHub

More ›